QUARANTEAM

Thinking beyond boundaries

Aboubacar

Doumbia

Michael

Bradshaw

Joane

Osei Owusu

Mohammed

Perves

Dhruv

Mehandiratta

Kimberly

Wang

What is our Mission?

page

Our Mandate

“To provide a strategic plan that optimizes the distribution of the COVID-19 vaccines in a

fair and unbiased manner”

Awareness

2 3

Prioritization

How do we prioritize

the remaining adults in

Phase 2 & 3 of the

Ontarian rollout?

How do we minimize

wastage of the

vaccine?

Distribution

How will the

population know

when to receive

vaccines?

Current Covid-19 vaccine distribution in Ontario

page

11,344 vaccines have

been wasted in 2 weeks

A Covid Vaccine produced

by Pfizer cost $19.50

$221,208

Underlying medical conditions at risk to Covid-19

page

Pop Size

Hypertension

Diabetes

Cardiovascular

Disease

Chronic

Obstructive

Lung

Disease

Cancer

Chronic

Kidney

191

30%

31.48%

1.68%

43.15%

23.36%

24.07%

7.41%

3.70%

137

23.36

1.46%

Awareness Plan

Create a marketing

campaign by utilizing

geo-targeted ads. Also

implement a health

passport

Our Strategy

Provincial Priority Level

Assess

Prioritization Plan

Create a detailed

plan ensuring the

people more at risks

gets vaccinated first

Inject

Scheduling Plan

Establish a website

where users can

schedule an

appointment online

or on the phone

Immunize

Making an impact to flatten the curve

Assess

Vaccine Prioritization Model - At the Individual Level

page

● Everyone in the province that we have the health data for will

be clustered and assigned a “provincial priority level” (PPL)

based on the cluster they belong to

● Clusters are assigned a “priority level” (PPL) ranging from 1 to 10

○ 1 Highest = Highest Priority | 10 = Lowest priority

○ Priority will be assigned based on the cluster's averages

of following factors (In order):

■ Age (old = higher priority)

■ Underlying health conditions

■ Income (low = higher priority)

● We will be using unsupervised machine learning models such

as K-means to cluster into k (we chose 10) clusters based on

the following numerical features:

● We will try to predict a PPL for users who have not been

assigned a PPL at the provincial level

● These users will have to fill in a questionnaire when they

○ Age

○ Obesity (1,0)

○ Hypertension (1,0)

○ Diabetes (1,0)

○ Other health conditions

○ Postal Code

○ Family Size

○ Essential Work

● Then they will be classified into the same 10 PPLs

● We will be using the following classification algorithms:

● Age

● Weight

● Income

● Essential

Worker

● MAX V02

● Blood Sugar

Level

● Cholesterol Level

● Blood pressure

● Postal Code Density

● Family Size

● Workplace Size (0

for WFH)

PPL 4

PPL

Note: Score is a measure of accuracy of the classifier on PPLs prediction on a

randomly generated mock dataset

Supervised Learning (For people without data)

Unsupervised Learning (For people with data)

Demographics Health Conditions Exposure & Spread

Classifier Score

KNeighborsClassifier 44%

SVC 40%

DecisionTreeClassifier 28%

RandomForestClassifier 32%

AdaBoostClassifier 40%

Classifier Score

GaussianNB 38%

LinearDiscriminantAnalysis 42%

QuadraticDiscriminantAnalysis 30%

NeuralNetwork

N/A

GradientBoostingClassifier 40%

K-means (K=10) Population Cluster’s Characteristics

page

Cluster Priority

Level (PPL)

0 (Blue) 6

1 (Orange) 10

2 (Red) 4

3 (Teal) 3

4 (Green) 7

5 (Yellow) 1

6 (Purple) 9

7 (Pink) 8

8 (Brown) 5

9 (Grey) 2

Order of Vaccination by Cluster

5, 9, 3, 2, 8, 0, 4, 7, 6, 1

Order determined by Age, Weight, and Income according to at-risk factors provided by

Public Health Ontario

Expert panel can look at individuals in clusters

page

Example

page

● If we had say the following 4 people register for vaccines at the ABC Shoppers

DrugMart On May 20, 2021

● And they had the following PPLs based on their characteristics

Name Age Obesity Hypertension Diabetes Income Family Size PPL Classifier

PPL

Abdul 66 1 0 1 Lower 7 2 N/A

Bob 23 0 0 0 Upper 3 9 N/A

Tyrone 34 0 0 1 Middle 5 6 N/A

Li 56 0 1 1 Middle 6 N/A 3

ABC Shoppers

DrugMart

Appointment Queue

Vaccine Distribution Model - At the PHU Level

page

Unsupervised Learning: K-means Clustering

PHUs are clustered using into k (we chose 5) clusters based on the

following Census and Covid features:

● Population age: 65+ yrs

● Average population age

● Population age: 0 - 17 yrs

● Population age: 18 - 64 yrs

● Population in 2016

● Active Covid Cases

● Resolved Covid Cases

● Covid Deaths

● Aboriginal Population

● Population in health services

● Population in education, law

and social, community and

govt services

● Population in trades, transport

and equipment operators and

related occupations

● Average total income of

households in 2015

● Population living in apartments

in buildings with 5+ storeys

Vaccines can be distributed based on the cluster each PHU

belongs to and the number of people in each PHU that belong to

the Population Clusters

Non-Political & Fair Distribution of Vaccines

AI Ethics: A Pillar of the Prioritization Model

page

General Concerns Mitigation

● Bias Variables such as income, race, or

religion

● Patient Confidentiality on Supervised

Machine Learning Data Collection Forms

○ Medical Conditions, Private Info, etc.

● Is the Data Collection, Transformation, and

Insight Storage Secure?

○ Will patients be informed of consent

and will they be promised the data

will not be shared other than its

intended use?

● Model Transparency

○ We avoid the use of “Black-Box”

Algorithms and will provide detailed

literature on how the prioritization model

came to a conclusion

○ Utilize Unbiased Training Data

● Prioritize the Investment of Top Notch Data

Security

● Develop and Implement Data Consent “Best

Practices” to Reassure the Public that Data

will be Used as Intended

● Allow a panel of Experts to view the

characteristics of each cluster, along with the

individuals within the cluster and assign the

priority level in a fair and unbiased manner

Inject

From the Clinics’ Perspective

page

Our vision for Scheduling

● Each PHU is further broken up into various

inoculation sites

○ Shoppers Drug Mart, Rexall, etc.

● Each PHU will receive a predetermined number of

vaccines per week, based on the prioritization

models

● Based on PPL data provided by the Provincial

Government, each PHU will be able to schedule

inoculations based on PPL level

What about Vaccine Wastage?

● Each subsidiary inoculation site for each PHU will

have access to the ledger of scheduling for its site

● If there are missed appointments, the site can start

contacting patients by PPL level scheduled for the

next day in order to minimize vaccine wastage

● Patients that missed their appointment will have to

re-engage with the Covid-19 web app to be

rescheduled

Web App

page

Immunize

Marketing Campaign

page

Health Passport

page

Given to everyone who

had taken the Covid-

19 vaccine

Public places are able to scan

the health passport to know if the

user has received the vaccine

Data will be stored on cloud

with end-to-end encryption

Bob Smith

Probability

Impact

People lying about their

medical condition

Significant amount of

people trying to register

on the web app at once

Someone tries to

duplicate the health

passport

Mitigations

Risks

Risks & Mitigations

Cross referencing their

health condition in the

provincial database

Integrate a time ticket

Each health passport will

be given an unique

identifier

Thank you!